diff --git a/README.md b/README.md index 44c893a..fe13fe3 100644 --- a/README.md +++ b/README.md @@ -1,263 +1,166 @@ -# LabNote +# CapstanAI LabNote -#### *A lightweight project notebook for keeping AI-assisted work coherent across sessions, tools and time.* +#### *A simple, file-based ledger for AI-assisted work that outlives the chat.* + +CapstanAI is the project name; LabNote is its practical, file-based ledger. **English** | [简体中文](locales/zh-CN/README.md) -

- CapstanAI LabNote raven perched on a lab notebook with a pearl and fountain pen +

+ Apache License 2.0 + No LabNote runtime

-LabNote helps you keep track of long-running projects that involve AI. - -If a project outlives one chat, moves between different AI tools, gets picked up again months later, or simply becomes too complicated to hold in your head, LabNote gives the work somewhere stable to live. - -It's a lightweight, repo-based notebook and coordination layer built from ordinary files. AI sessions can leave behind useful work, provenance, decisions, critiques, handoffs and next steps, so the next session does not have to start from a blank room. - -You can use it with ChatGPT, Codex, Claude Code, local models, browser AIs, coding agents, or any other AI that can work with ordinary repository files. - -## What can you use LabNote for? - -LabNote is deliberately general-purpose. - -**Software and technical projects** -Coordinate planning, implementation, testing, review and handoffs across different AI tools. - -**Research and investigations** -Keep sources, findings, competing hypotheses, evidence and synthesis organised across separate sessions. - -**Writing and creative work** -Carry outlines, drafts, critiques, continuity notes and editorial decisions through a long project. - -**Casework and life admin** -Keep chronology, correspondence, evidence, drafts, decisions and next actions together. - -**Learning and academic work** -Preserve research notes, questions, explanations, progress and tutor-session continuity. - -**Dormant projects** -Come back after weeks or months and ask a fresh AI to bring you up to speed without rebuilding the whole context by hand. - -Those are just a few examples. - -If the work lasts longer than one chat, benefits from more than one AI session, or would be a drag to reconstruct from memory, then LabNote probably has a use for it. - -## Why I made it - -Fact is, I got sick and tired of the context window for chats filling up long before I'd even finished thrashing out an idea for a project. Sure, you can generate a handoff sheet for the next AI, but that means asking it to synthesise the whole chat, which means some finer points and even entire avenues of thought can be lost. -At the same time, I'd keep catching myself coming up with development ideas on an unrelated chat, or even on a chat with another model. This meant that things would get scattered and I'd lose track to the point that sometimes I'd even abandon a project and start from scratch. - -To try and keep the sprawl under control, I burned time, tokens and compute, passing around progress reports, handover sheets, documents, critiques and reminders - all just to keep everything on the same page. -AI platforms have tried to address continuity with things like project areas, canvases and branching chats. They help, but they still don't do the job for me. -I needed something simpler and more intuitive; something that would let one AI chat flow naturally into another, and would also let me keep track of projects without things turning into a full-on second-brain. -Hence LabNote. -What started as a fix for my own terrible project memory turned into something considerably more useful. - -Fundamentally, using LabNote, the AI is mostly relegated to a low-token, low-compute archivist and coordinator. It's told what to read, where to put things, what to record and when to stop. - -LabNote is my daily driver now. I make a new one for most substantial projects, and I even use it to dust down old work that has been sitting untouched for months. - -## The basic idea +

+ CapstanAI LabNote raven perched on a lab notebook with a pearl and fountain pen +

Most AI work starts in a blank room. -A fresh chat does not know what happened before. One model may not know what another decided. A coding agent may solve its narrow task but miss the wider project. Useful reasoning can disappear into an old conversation nobody wants to reread. - -LabNote gives your AI a room with labels on the drawers. - -A session can enter, find the relevant work, contribute what it needs to contribute, leave a traceable record, and stop. +A fresh chat often does not know what happened before. One model may not know what another decided. A coding agent may finish its own task without seeing the wider project. Useful work gets stranded in old conversations that nobody wants to reconstruct by hand. -The human remains in charge. - -A typical trail is: - -```text -packet → response → review → decision -``` +**LabNote gives your AI a room with labels on the drawers.** -No daemon. -No database. -No repo-resident agent. -No model API keys. -No shared-memory theatre. +It is a lightweight project notebook built from ordinary files. It gives selected project material a stable home, so a new session can find the thread instead of starting again. -Just ordinary files arranged so useful project knowledge survives the chat that created it. +Set it up once, point each AI at the same entrance, and let the ledger carry the continuity. -## How it works +With suitable access to the files, LabNote works with browser AIs, coding agents and local models. If an AI cannot reach the files itself, it can still guide you through getting a copy and setting it up. -That simple trail expands into a repeatable handoff: +## What LabNote does well -```text -Human or AI creates a packet -↓ -Packet goes into the right inbox -↓ -A receiving AI session reads only what it needs -↓ -The receiving session writes a response -↓ -The response is reviewed -↓ -The decision is recorded -``` +You may already be running a rough version of this yourself: pasting summaries between chats, writing handoff notes, keeping scattered folders, bookmarking old conversations, and trying to remember which AI was told what. -A good handoff carries enough provenance and status to show where the work came from, what happened to it, and what should happen next. +LabNote turns that repeated coordination work into a visible route through the project. It can be a task list, a handoff ledger, a small working library, an audit trail, or a deliberate drop point between sessions. -Once LabNote is set up, you can ask an AI or agent to do things like: +It gives a project a durable place to: -```text -Put this document in my LabNote repo and cross-reference it with [document name]. -``` +- keep the important tasks and decisions, with their sources and next actions; +- pass work between sessions without repeating the whole brief; +- keep critique and review attached to the work they concern; and +- return after a gap with a clear trail instead of a pile of old chats. -```text -Leave a critique for [AI/Agent name] about [document name]. -``` +LabNote is an addition to your existing tools, not a replacement for them. It sits alongside ordinary folders, version control, and project applications. You decide what belongs in the record and what requires review; your AI tools help do the work, while the workspace holds the trail. -```text -Register this report, tag it properly, and leave a short completion note. -``` +Give each meaningful project its own named LabNote copy. Start with a few tasks, one handoff or a small document trail; let it grow into a fuller record only when the work calls for it. -LabNote gives the AI enough structure to act, and enough constraint to stop guessing. +When a chat ends, a tool changes, or a project goes quiet, the files remain. Kept under normal version control and backup, they give the next session a way back into the work, and a map for rebuilding it. ## Get LabNote -If you are already talking to an AI, the easiest starting point is: +> **Already looking at your own copy?** You have LabNote already—start at +> [`AI_ENTRYPOINT.md`](AI_ENTRYPOINT.md). + +If you are already talking to an AI, this is the easiest place to begin. Copy and paste: ```text Help me set up CapstanAI LabNote for this project. First work out what access you have, then recommend the simplest safe route for me. I want you to explain any terminal commands before I run them. ``` -Common routes: +Or make your own copy directly: -| Environment | Suggested route | +| Where you want it | What to do | | --- | --- | -| Browser AI + terminal available | AI-guided local clone, then connect to a private GitHub repo if wanted | -| Browser AI without terminal access | GitHub **Use this template** route | -| Coding agent or terminal-capable AI | Agent clones/copies and configures the workspace, subject to operator approval | -| Local-only project | Clone or copy locally; no remote required | - -See [`docs/ACQUISITION.md`](docs/ACQUISITION.md) for exact beginner-friendly steps and terminal guidance. +| **In your own GitHub repository** | [Use this template](https://github.com/Wonderforge-Lab/CapstanAI-LabNote/generate), then choose the name and visibility of your new repository. | +| **On your own computer** | Clone or download the repository. A remote GitHub copy is optional. | -The acquisition method does not change how LabNote works after the copy exists. +> **Keep live work out of this public repository.** The public CapstanAI LabNote repository is the source scaffold. Put private project material in your own private, controlled or local copy. -### Public source, live workspace +Once you have your copy: -The canonical CapstanAI LabNote repository is public so people can inspect and acquire the scaffold. A copied workspace may be private, public, local-only, or otherwise controlled. Portable LabNote workflow files do not depend on a fixed repository owner, slug, or visibility. +1. Give your AI [`AI_ENTRYPOINT.md`](AI_ENTRYPOINT.md) as its starting point. +2. Confirm the workspace context if it asks. +3. Let it follow the LabNote lobby from there. -If the current workspace is public or reference-only, do not deposit private runtime material there. Controlled live workspaces follow the runtime and storage rules linked below. +[Choose the right setup route](docs/ACQUISITION.md), or [follow the first-use walkthrough](docs/quickstart.md). -## First use +## A working picture -1. Acquire the workspace. -2. Give your AI `AI_ENTRYPOINT.md` as its starting point. -3. Confirm the workspace context if the AI asks. -4. Let the AI follow the LabNote lobby from there. - -No repository rename or LabNote identity-file edit should be required before first use. You do not need to learn the whole filing system before using it. +```text +lobby/ ← every AI session starts here +datadrops/ ← source material and work go in +responses/ ← critiques, replies and signoffs come back +registry/ ← what happened, when and by whom +``` -See [`docs/quickstart.md`](docs/quickstart.md) for the longer walkthrough. +A session enters, reads the relevant route, contributes what it needs to contribute, leaves a traceable record and stops. The human remains in charge. -## Why it helps +A typical trail is **packet → response → review → decision**. -* **Preserve continuity** across fresh chats, models, tools, coding agents, and dormant projects. -* **Cut coordination admin** by letting AI sessions leave structured work for each other instead of making the human relay everything by hand. -* **Keep work auditable** through labelled packets, provenance, responses, reviews, visit records, and signoffs. -* **Bound AI behaviour** with deterministic entry, fixed routes, ask-gates, stop-points, and human review. -* **Stay repo-native and low-bloat** so any capable AI that can read and write ordinary repo files can participate without a daemon, database, or hidden runtime. +Once LabNote is set up, you can say things like: -## Why not just use `AGENTS.md`? +```text +Put this document in my LabNote and cross-reference it with [document name]. -`AGENTS.md`, `CLAUDE.md`, and similar context files are useful. LabNote can work alongside them, but it is solving a different problem. +Leave a critique for [AI or agent name] about [document name]. -A typical agent context file usually tells an AI about a repository: how to run tests, where key files live, what style to follow, and what commands to use. That can help, but it can also turn into a long instruction blob as the project grows. +Register this report, tag it properly, and leave a short completion note. +``` -LabNote is not just a bigger context file. It gives the AI a structured workflow: where to enter, what to read first, where to deposit work, how to tag it, how to sign off, how to hand work onward, and when to stop and ask the human. +The AI gets enough structure to act, enough context to orient itself, and clear stopping points when a human decision is needed. The included validation workflow runs on pushes and pull requests. It checks registry records and provenance/path rules, generated views and Markdown links, locale invariants and freshness, bridge configuration, and tag-promotion separation. -| Ordinary agent context file | LabNote | -| --- | --- | -| Tells the AI about the repo | Gives the AI a route through the work | -| Can become a large instruction blob | Uses nested, role-specific instructions | -| Often focuses on task execution | Also handles handoffs, review, provenance, and stop-points | -| May increase exploration and context load | Keeps work bounded through packet routes and ask-gates | -| Usually lives as one file | Uses a small repo structure, templates, registry records, and signoffs | -| Helps one agent orient itself | Helps many AI sessions coordinate over time | +[See how handoffs move through review and decision](docs/review_workflow.md). -`AGENTS.md` can tell an AI what kind of project it is in. +## Why clone it instead of making a folder? -LabNote tells the AI how to behave inside the project. +You absolutely can make a folder, keep a to-do list and ask an AI to leave better notes. For a short project, that may be all you need. -## Built around five principles +LabNote is the version you do not have to keep rebuilding. It gives every session the same place to begin, a small method for leaving work behind, and a visible record that still makes sense when you return later or switch tools. -1. **Human-held authority** - LabNote supports the human-in-the-loop. It does not replace them. +You do not have to use every part of it. The more careful routes are already there when you need review, provenance, auditability or a clean way to pass work on. Until then, it can sit quietly beside the tools you already use. -2. **Deterministic entry** - Every AI enters through the lobby, reads the same rules, and follows the same route. +## Why LabNote stays small -3. **Bounded action** - The AI gets fixed targets, allowed paths, stop conditions, and ask-gates. +LabNote does not try to save everything as it happens. Not every message belongs in a durable project record. Instead, it preserves the parts you choose to carry forward: the source, the decision, the critique, the handoff and the next action. -4. **Traceable work** - Documents, critiques, tags, decisions, handoffs, and signoffs leave a clear trail. +That keeps the ledger smaller, clearer and easier to inspect. It avoids needless rereading, resummarising and context loading when a later session only needs the useful trail. -5. **Growth without bloat** - LabNote can grow into richer workflows without requiring a local daemon, database, model install, or repo-resident agent. +The rails do not make a model smarter, and they cannot make one infallible. They make routine coordination work smaller and clearer: a known entrance, a limited reading route, clear write targets, and defined points to stop and ask. -## Where to go next +LabNote itself needs: -LabNote is the ledger, not the warehouse. Keep the workspace focused on small, inspectable records and approved references to heavier material. Do not store credentials, private keys, or tokens in LabNote. +- no daemon; +- no database; +- no hosted runtime; +- no repo-resident agent; +- no model API keys. -| Area | Need | Read | -| --- | --- | --- | -| **Start** | Set up LabNote | [`docs/ACQUISITION.md`](docs/ACQUISITION.md) | -| **Start** | Beginner walkthrough | [`docs/quickstart.md`](docs/quickstart.md) | -| **Start** | Give an AI its starting instructions | [`AI_ENTRYPOINT.md`](AI_ENTRYPOINT.md) | -| **Work** | Make a normal deposit | [`lobby/ROUTINE_DEPOSIT_QUICKSTART.md`](lobby/ROUTINE_DEPOSIT_QUICKSTART.md) | -| **Work** | Understand visitor/session identity | [`docs/visitor_lobby_model.md`](docs/visitor_lobby_model.md) | -| **Work** | Route messages between sessions | [`docs/message_routing_model.md`](docs/message_routing_model.md) | -| **Work** | Review a response or decision | [`docs/review_workflow.md`](docs/review_workflow.md) | -| **Reference** | Branch and PR rules | [`docs/BRANCH_HYGIENE.md`](docs/BRANCH_HYGIENE.md) | -| **Reference** | Registry records and paths | [`docs/REGISTRY_RECORDS.md`](docs/REGISTRY_RECORDS.md) | -| **Reference** | Storage rules | [`docs/storage_policy.md`](docs/storage_policy.md) | -| **Reference** | Document and binary deposit rules | [`docs/DOCUMENT_DEPOSIT_POLICY.md`](docs/DOCUMENT_DEPOSIT_POLICY.md) | +No shared-memory theatre. -Detailed runtime rules live in those canonical files rather than being duplicated here. +Just ordinary files, clear routes and human-held authority. The ledger is right there in the files, where you can inspect it yourself. -## Acknowledgements +[See how LabNote fits alongside context files and model memory](docs/WHY_LABNOTE.md). -CapstanAI - LabNote is human-led. Recent language-layer and verification work was assisted by the following AI collaborators: +## Where it earns its keep -- **OpenAI Codex** — implementation support, verification design, and repository review. -- **DeepSeek** — independent Simplified-Chinese language, cultural, and adversarial review. +**Research, investigations and casework** +Keep sources, evidence, competing explanations, chronology and decisions together across separate sessions. -These acknowledgements describe assisted contributions. Human maintainers retain authorship, decision authority, and release control. +**Software and technical work** +Carry plans, implementation notes, tests, reviews and handoffs between different tools and agents. -## Status +**Writing, study and creative projects** +Preserve outlines, drafts, critiques, open questions and editorial choices without rebuilding the brief every time. -Latest published scaffold release: `v0.3.2 - Locale Route Freshness`. +**Long-running or dormant projects** +Return after weeks or months and give a fresh AI a reliable place to find what mattered and what should happen next. -This patch release: +The subject does not matter much. If the work lasts longer than one chat, moves between tools, or would be a nuisance to reconstruct from memory, LabNote can give it a durable spine. -* corrects stale zh-CN CSV guidance on live visitor routes; -* aligns the zh-CN visitor-lobby model with the canonical routing and registration rule; -* adds CI checks for protected locale invariants and for unreviewed changes to paired English sources; +## Why I made it -It leaves Registry Contract v1, schemas, runtime paths, machine fields, statuses, permissions, generated views, and the shared protocol unchanged. +Fact is, I got sick and tired of chat context windows filling up before I'd even finished thrashing out an idea. -`v0.3.0 - Registry v1 and Simplified Chinese Revision` remains the feature baseline beneath these patch releases. +Handoff sheets helped, but each new synthesis risked losing finer points or whole avenues of thought. Ideas would appear in unrelated chats or conversations with another model. Progress reports, critiques and reminders multiplied, and sometimes the project became scattered enough that I lost the thread and started again. -CapstanAI may later grow a richer deterministic layer, along with relay, vault, and protocol modules. LabNote begins as the simplest useful ledger. +Project areas, canvases and branching chats all help, but I wanted something simpler: a way for one AI session to flow naturally into another without turning the project into a full-on second brain. -```text -packets, provenance, replies, and decisions -``` +Hence LabNote. What began as a fix for my own terrible project memory became my daily driver. I now make one for most substantial projects, and use it to dust down work that has been sitting untouched for months. -## Motto +At heart, it turns the AI into a low-overhead archivist and coordinator: told what to read, where to put things, what to record and when to stop. -```text -Mind the gap. Mark the crossing. -``` +## Licence -## License +[Apache License 2.0](LICENSE) · [CapstanAI LabNote source releases and version history](https://github.com/Wonderforge-Lab/CapstanAI-LabNote/releases) -Apache License 2.0. +> **Mind the gap. Mark the crossing.** diff --git a/docs/PROJECT_PROVENANCE.md b/docs/PROJECT_PROVENANCE.md new file mode 100644 index 0000000..d716079 --- /dev/null +++ b/docs/PROJECT_PROVENANCE.md @@ -0,0 +1,42 @@ +# Project provenance + +CapstanAI LabNote is human-led. Human maintainers retain authorship, decision +authority, repository control, and release authority. + +The project has also used managed AI assistance for development, implementation, +testing, review, localisation, and documentation. The roles below describe that +assistance transparently; they do not transfer responsibility away from the +human maintainers. + +## Managed AI contributions + +| Contributor | Role in the project | +| --- | --- | +| ChatGPT (“Cal”) | Project design, documentation development, implementation coordination, and review | +| Claude | Independent critique, testing, and editorial review | +| OpenAI Codex | Repository implementation, validation design, and technical review | +| Claude Code | Implementation and test assistance | +| DeepSeek | Simplified-Chinese translation assistance and linguistic, cultural, and adversarial review | + +## What this record means + +This is a project-wide provenance record. It describes the kinds of assistance +used while the project was developed and reviewed. + +It is not an employment record, an agency relationship, an authorship transfer, +or a claim that an AI system holds responsibility for a project decision, +repository change, or release. + +## Revision records + +Specific releases, pull requests, review records, and revision notes retain +their own local history. They may name the work or review relevant to that +revision without needing to repeat this full project-wide record. + +## Localisation + +English is the canonical protocol language. Locale documentation may explain +the same project facts in a natural local form while preserving the relevant +paths, machine fields, and protocol boundaries. + +[Return to the documentation index](README.md). diff --git a/docs/README.md b/docs/README.md new file mode 100644 index 0000000..9fd475b --- /dev/null +++ b/docs/README.md @@ -0,0 +1,46 @@ +# CapstanAI LabNote documentation + +The repository README helps you decide whether LabNote fits your work and get +a copy. This page helps you find the exact guidance you need once you have +one. + +**English** | [简体中文](../locales/zh-CN/docs/README.md) + +## Start + +- [Choose a setup route](ACQUISITION.md) — use the repository as a GitHub + template, clone or copy it locally, or ask an AI to guide you through the + safest route it can support. +- [Follow the first-use walkthrough](quickstart.md) — begin in a copied + workspace without learning the entire structure first. +- [Start an AI at the entrypoint](../AI_ENTRYPOINT.md) — the controlled + reading route for a session working inside LabNote. + +## Work + +- [Make a routine deposit](../lobby/ROUTINE_DEPOSIT_QUICKSTART.md) — place a + normal piece of work in the right route with the minimum useful record. +- [Understand review and decisions](review_workflow.md) — see how a response + becomes a reviewed outcome. +- [Understand visitor and session identity](visitor_lobby_model.md) — learn + what session records mean in a live workspace. +- [Understand message routing](message_routing_model.md) — follow the model + for messages between sessions or roles. +- [Understand registry records](REGISTRY_RECORDS.md) — see what the durable + register records are for. + +## Keep the workspace safe + +- [Read the storage policy](storage_policy.md) — keep the ledger small, + inspectable, and appropriate to its visibility. +- [Read the document-deposit policy](DOCUMENT_DEPOSIT_POLICY.md) — handle + documents and binaries without turning LabNote into a general file dump. +- [Read the security guidance](../SECURITY.md) — do not put credentials, + tokens, keys, or unintended private material into the wrong workspace. + +## Understand the project + +- [LabNote alongside context files and model memory](WHY_LABNOTE.md) — how a + visible project ledger complements those two kinds of AI support. +- [Project provenance](PROJECT_PROVENANCE.md) — how human maintainers and + managed AI assistance contributed to the project. diff --git a/docs/WHY_LABNOTE.md b/docs/WHY_LABNOTE.md new file mode 100644 index 0000000..3814905 --- /dev/null +++ b/docs/WHY_LABNOTE.md @@ -0,0 +1,43 @@ +# LabNote alongside context files and model memory + +LabNote keeps selected project continuity in visible files. It complements a +model's context window and repository instruction files; it does not replace +them. + +## Model context is not project continuity + +A long context window can help a model read more in one sitting. It does not, +by itself, decide what should remain important after the session, show the next +tool why a decision was made, or create a selective record that a human can +inspect. + +LabNote does not enlarge a model’s native memory or silently capture your +conversations. A human or an AI session deliberately writes the record. That is +why the trail can be checked, corrected, reviewed, and carried to another +tool. + +Context length is useful. Project continuity is a separate job. + +## Context files set local rules + +Files such as `AGENTS.md` or `CLAUDE.md` are useful ways to tell an AI about a +repository: where important files are, how to run tests, and what local rules +apply. + +LabNote complements them. Its job is to route ongoing project work: where a +session begins, what it should read, where it may leave work, how that work is +reviewed, and when the session should stop and ask. + +A context file tells an AI what kind of repository it is in. LabNote gives it a +route through the work happening there. + +## Use the smallest useful amount + +If a one-shot answer is enough, use the best tool available and get on with it. +If a plain folder is enough, use a plain folder. + +Use LabNote when you want a visible working trail that can grow with the +project without becoming a hidden service or a second brain. Start with the +smallest record that will genuinely help the next session. + +[Return to the documentation index](README.md). diff --git a/locales/zh-CN/README.md b/locales/zh-CN/README.md index 8a34999..9ccb560 100644 --- a/locales/zh-CN/README.md +++ b/locales/zh-CN/README.md @@ -1,270 +1,166 @@ -# LabNote +# CapstanAI LabNote -#### *一个轻量级项目笔记本，让 AI 辅助工作跨会话、跨工具、跨时间保持连贯。* +#### *一个简洁、基于文件的项目工作台账（ledger），让 AI 辅助工作在聊天结束后仍能延续。* + +CapstanAI 是这个项目的名称；LabNote 是其中实用、基于文件的工作台账。 [English](../../README.md) | **简体中文** -

- CapstanAI LabNote 乌鸦栖在实验笔记本上，旁边有一颗珍珠和一支钢笔 +

+ Apache License 2.0 + 无需 LabNote 运行时

-LabNote 帮你把那些需要长期与 AI 协作的项目整理清楚。 - -如果一个项目活得比一次聊天更久，要在不同 AI 工具之间来回切换，几个月后还会重新捡起来，或者复杂到已经不适合全塞在脑子里，LabNote 就给这些工作一个稳定的落脚处。 - -它是一个轻量级、基于代码仓库的笔记与协调层，由普通文件组成。不同 AI 会话可以留下有用的工作成果、溯源信息（provenance）、决定、批评意见、交接内容和下一步，让后来的会话不必每次都从一间空屋子开始。 - -你可以把它和 ChatGPT、Codex、Claude Code、本地模型、浏览器 AI、编程智能体，或者任何能够读写普通仓库文件的 AI 一起使用。 - -## LabNote 可以拿来做什么？ - -LabNote 本来就是按通用用途设计的。 - -**软件与技术项目** -在不同 AI 工具之间协调规划、实现、测试、审阅和交接。 - -**研究与调查** -把来源、发现、相互竞争的假设、依据材料或证据，以及综合分析，跨多个独立会话整理清楚。 +

+ CapstanAI LabNote 乌鸦栖在实验笔记本上，旁边有一颗珍珠和一支钢笔 +

-**写作与创作** -让提纲、草稿、批评意见、连续性笔记和编辑决定能够贯穿整个长期项目。 +许多 AI 工作都从一间空屋子开始。 -**个案工作与生活事务** -把时间线、往来通信、证据、草稿、决定和下一步行动放在一起。 +新的聊天不知道此前发生了什么。一个模型未必知道另一个模型做过什么决定。编程智能体可能完成了自己的任务，却看不到项目的整体方向。有价值的工作则沉在旧对话里，没人愿意靠手工重新拼起来。 -**学习与学术工作** -保留研究笔记、问题、解释、进度，以及不同辅导会话之间的连续性。 +**LabNote 给 AI 的，是一间抽屉都贴好标签的屋子。** -**搁置中的项目** -几周或几个月后回来时，可以让一个全新的 AI 直接帮你恢复上下文，而不是手工把整套背景重新拼一遍。 +它是一个由普通文件组成的轻量级项目笔记本，为你选择保留的项目材料提供稳定的归处，让新会话能找到线索，而不是从头开始。 -这些只是几个例子。 +设置一次，让每个 AI 都从同一个入口开始；随后由这份台账承载项目的连续性。 -只要一项工作比一次聊天更长，需要多个 AI 会话协作，或者仅凭记忆重建起来很麻烦，LabNote 多半就能派上用场。 +只要具备相应的文件访问能力，LabNote 就能与浏览器 AI、编程智能体和本地模型配合使用。即使 AI 本身无法访问这些文件，它仍可引导你取得副本并完成设置。 -## 我为什么做它 +## LabNote 擅长做什么 -说实话，我受够了一件事：项目里的想法还没琢磨明白，聊天的上下文窗口倒先塞满了。你当然可以给下一个 AI 做一份交接摘要，但那意味着让它重新梳理整场聊天，而在这个过程中，一些细节，甚至整条思路，都可能被漏掉。 -与此同时，我又总会在完全不相干的聊天里冒出某个开发点子，有时甚至是在另一个模型的对话里。结果就是东西越散越开，我自己也越来越跟不上。偶尔乱到一定程度，我干脆把项目扔掉，从头再来。 +你可能早就在用粗略的方式做这件事：在聊天之间粘贴摘要、写交接说明、保留零散文件夹、收藏旧对话，或努力记住曾向哪个 AI 交代过什么。 -为了压住这种蔓延，我花掉了不少时间、token 和算力，在不同地方来回传进度报告、交接单、文档、批评意见和提醒，只为了让大家还在看同一张地图。 -AI 平台也一直在尝试解决连续性问题，比如项目空间、画布、聊天分支。它们有帮助，但对我来说仍然不够用。 -我需要的是更简单、更符合直觉的东西：让一个 AI 对话能自然地流进另一个，也让我自己能跟得住项目，又不至于把系统变成一个庞大的“第二大脑”。 -于是就有了 LabNote。 -最初，它只是用来补救我糟糕的项目记忆，后来却长成了一个有用得多的东西。 - -从根本上说，在 LabNote 里，AI 大多数时候只是一个低 token、低算力的档案员和协调员。系统会告诉它该读什么、东西放哪里、需要记录什么，以及什么时候该停。 +LabNote 把这些反复出现的协调工作变成一条看得见的项目路径。它可以是待办事项清单、交接台账、小型工作资料库、审计轨迹，或会话之间有意设置的投递点。 -LabNote 现在已经是我的日常工具了。大多数稍微像样一点的项目，我都会新建一个；那些放了几个月没碰的旧工作，我也会拿它出来掸掸灰，再接着做。 +它为一个项目提供一个可以长期保留的位置，用来： -## 基本思路 +- 保存重要的任务和决定，以及相应的来源材料和下一步； +- 在会话之间交接工作，而无需每次重述完整背景； +- 让批评意见和审阅与其对应的工作保持关联；以及 +- 在间隔一段时间后回来时，顺着清晰的轨迹继续，而不是翻找一堆旧聊天。 -大多数 AI 工作，都是从一间空屋子开始的。 +LabNote 是对现有工具的补充，不会取代它们。它可以与普通文件夹、版本控制和项目应用并存。由你决定哪些内容进入记录、哪些需要审阅；AI 工具协助完成工作，工作区则保存轨迹。 -新的聊天不知道之前发生过什么。一个模型未必知道另一个模型做过什么决定。编程智能体可能把眼前的小任务解决得很好，却错过整个项目的方向。有价值的推理也可能就此沉进一场没人想再翻的旧对话里。 +每个重要项目都可以有一个属于自己的 LabNote 副本。可以从几项任务、一份交接或一条简短的文档线索开始；只有在工作确实需要时，再逐步扩展成更完整的记录。 -LabNote 给 AI 的，是一间抽屉都贴好标签的屋子。 +当聊天结束、工具更换或项目沉寂时，文件仍然存在。配合正常的版本控制和备份，它们能让下一次会话重新进入工作，并为重建工作提供路线图。 -一个会话可以进来，找到相关工作，完成它需要完成的部分，留下可追踪的记录，然后停下。 +## 获取 LabNote -人类始终掌握主导权。 +> **已经在查看你自己的副本？** 你已经拥有 LabNote——请从 +> [`AI_ENTRYPOINT.md`](../../AI_ENTRYPOINT.md) 开始。 -一条典型路径是： +如果你正在与 AI 对话，最容易的起点是复制并粘贴： ```text -packet → response → review → decision -``` - -也就是： - -```text -工作包 → 回复 → 审阅 → 决定 +请帮我为这个项目设置 CapstanAI LabNote。先判断你目前具备哪些访问能力，再向我推荐最简单、最安全的方式。如果需要我运行任何终端命令，请先解释命令的作用。 ``` -没有守护进程。 -没有数据库。 -没有常驻仓库的智能体。 -没有模型 API 密钥。
-也不演“共享记忆”这出戏。 +或者直接创建自己的副本： -只有一些普通文件，被安排成一种让项目知识活得比生成它的那场聊天更久的方式。 +| 希望放在哪里 | 该怎么做 | +| --- | --- | +| **自己的 GitHub 仓库中** | [Use this template](https://github.com/Wonderforge-Lab/CapstanAI-LabNote/generate)，然后为新仓库选择名称和可见性。 | +| **自己的电脑上** | 克隆或下载该仓库；是否使用远程 GitHub 副本由你决定。 | -## 它是怎么工作的？ +> **请勿在这个公开仓库中放入实际工作。** 公开的 CapstanAI LabNote 仓库只是源框架；请把私有项目材料放在你自己的私有、受控或本地副本中。 -上面那条简单路径，会展开成一套可以反复使用的交接流程： +取得副本后： -```text -人类或 AI 创建一个工作包（packet） -↓ -工作包进入正确的收件箱 -↓ -接收它的 AI 会话只读取真正需要的内容 -↓ -接收会话写出回复 -↓ -回复接受审阅 -↓ -决定被记录下来 -``` +1. 把 [`AI_ENTRYPOINT.md`](../../AI_ENTRYPOINT.md) 作为 AI 的起点交给它。 +2. 如果它要求确认工作区环境，请予以确认。 +3. 然后让它从 LabNote 的入口区（lobby）继续。 -一次好的交接，会带上足够的溯源信息和状态，让后来的人或 AI 看得出这项工作从哪里来、发生了什么，以及下一步该做什么。 +[选择合适的设置方式](docs/ACQUISITION.md)，或[按首次使用流程开始](docs/quickstart.md)。 -LabNote 设置好以后，你可以请 AI 或智能体做这样的事： +## 工作方式一览 ```text -把这份文档放进我的 LabNote 仓库，并和 [文档名称] 建立交叉引用。 +lobby/ ← 每个 AI 会话从这里开始 +datadrops/ ← 来源材料和工作放在这里 +responses/ ← 批评意见、回复和签退记录在这里返回 +registry/ ← 记录发生了什么、何时发生以及由谁完成 ``` -```text -针对 [文档名称]，给 [AI/智能体名称] 留一份批评意见。 -``` +一个会话进入后，读取相关路径，完成自己需要贡献的部分，留下可追溯的记录，然后停止。人类始终掌握主导权。 -```text -登记这份报告，正确打上标签，并留一条简短的完成记录。 -``` - -LabNote 给 AI 足够的结构去行动，也给它足够的边界，让它知道什么时候不能靠猜。 - -## 获取 LabNote +一条典型轨迹是 **packet → response → review → decision**；对应地，它是工作包（packet）→ 回复（response）→ 审阅（review）→ 决定（decision）。 -如果你已经在和一个 AI 对话，最简单的起点就是把下面这段话交给它： +设置好 LabNote 后，你可以这样对 AI 说： ```text -请帮我为这个项目设置 CapstanAI LabNote。先判断你目前具备哪些访问能力，再向我推荐最简单、最安全的方式。如果需要我运行任何终端命令，请先解释命令的作用。 -``` - -常见方式： - -| 环境 | 建议方式 | -| --- | --- | -| 浏览器 AI（可使用终端） | 由 AI 引导在本地克隆；如有需要，再连接到私有 GitHub 仓库 | -| 浏览器 AI 无法使用终端 | 使用 GitHub 的 **Use this template** | -| 编程智能体或可使用终端的 AI | 在操作者（human operator）批准的前提下，由智能体克隆/复制并配置工作区 | -| 仅本地项目 | 直接在本地克隆或复制；不需要远程仓库 | - -具体的入门步骤和终端说明，请参阅简体中文指南 [`docs/ACQUISITION.md`](docs/ACQUISITION.md)。 - -一旦工作区副本已经存在，采用哪种获取方式并不会改变 LabNote 后续的工作方式。 - -### 公共源仓库与实际工作区 - -CapstanAI LabNote 的基准（canonical）仓库是公开的，方便任何人检查和获取这套框架。复制出来的工作区可以是私有的、公开的、只存在于本地的，或者处于其他受控状态。LabNote 的可移植工作流文件不依赖固定的仓库所有者、仓库名或可见性。 - -如果当前工作区是公开的，或者仅供参考，请不要把私密的运行期材料投递进去。受控工作区（controlled live workspace）则应遵循下面链接的运行与存储规则。 - -## 第一次使用 - -1. 获取工作区。 -2. 把 `AI_ENTRYPOINT.md` 作为 AI 的起点交给它。 -3. 如果 AI 要求确认当前工作区环境，请确认。 -4. 然后让 AI 从 LabNote 的入口区（lobby）继续按规则往下走。 - -第一次使用前，你不需要先修改仓库名称，也不需要修改 LabNote 的身份文件。你更不必先学完整套归档系统，就可以开始使用。 - -更完整的流程请参阅简体中文指南 [`docs/quickstart.md`](docs/quickstart.md)。 +把这份文档放进我的 LabNote，并和 [文档名称] 建立交叉引用。 -## 它为什么有用？ +针对 [文档名称]，给 [AI 或智能体名称] 留一份批评意见。 -* **保留连续性**：让新聊天、不同模型、不同工具、编程智能体和搁置中的项目仍然能接得上。 -* **减少协调事务**：让各个 AI 会话自行留下结构化工作，而不是所有东西都要靠人工转递（human relay）。 -* **让工作可审计**：通过带标签的工作包、溯源信息、回复、审阅、访问记录和签退记录（signoff），留下清楚的轨迹。 -* **约束 AI 行为**：通过确定性入口、固定路由、询问节点（ask-gate）、停止点和人工审阅，为 AI 的行动划定边界。 -* **保持仓库原生、控制膨胀**：任何能读写普通仓库文件的 AI 都能参与，不需要守护进程、数据库或隐藏运行时。 - -## 为什么不直接用 `AGENTS.md`？ - -`AGENTS.md`、`CLAUDE.md` 和类似的上下文文件都很有用。LabNote 可以和它们一起使用，但解决的是另一类问题。 - -普通的智能体上下文文件，通常是在告诉 AI 这个仓库是什么：测试怎么跑、关键文件在哪里、遵循什么风格、可以用哪些命令。这样当然有帮助，但项目一大，它也可能慢慢长成一大团说明文字。 - -LabNote 并不是一份更大的上下文文件。它给 AI 的是一条结构化工作路径：从哪里进入、先读什么、工作投递到哪里、怎样打标签、怎样签退、如何把工作交给下一个会话，以及什么时候必须停下来询问人类。 +登记这份报告，正确打上标签，并留一条简短的完成记录。 +``` -| 普通智能体上下文文件 | LabNote | -| --- | --- | -| 告诉 AI 仓库是什么 | 给 AI 一条穿过整个工作的路径 | -| 可能越长越像一大团说明 | 使用分层、按角色划分的说明 | -| 通常关注任务执行 | 同时处理交接、审阅、溯源和停止点 | -| 可能增加探索范围和上下文负担 | 通过工作包路由和询问节点保持边界 | -| 通常就是一个文件 | 使用小型仓库结构、模板、登记记录和签退记录 | -| 帮一个智能体定位自己 | 帮多个 AI 会话跨时间协调 | +AI 获得足够的结构来行动，足够的背景来定位自己；当需要人类决定时，它也有明确的停止点。随附的验证工作流会在每次推送和拉取请求时运行。它会检查登记记录及其溯源信息/路径规则、生成视图和 Markdown 链接、本地化不变量与源文件新鲜度、bridge 配置，以及标签晋级隔离。 -`AGENTS.md` 可以告诉 AI：你现在身处什么项目。 +[查看交接如何经过审阅并形成决定](docs/review_workflow.md)。 -LabNote 告诉 AI：你在这个项目里应该怎么行动。 +## 为什么要克隆它，而不是自己建个文件夹？ -## 五个基本原则 +当然可以自己建一个文件夹，保留待办事项清单，再要求 AI 写更好的笔记。对于短项目，这可能已经足够。 -1. **最终决定权由人类掌握** - LabNote 支持人在回路（Human-in-the-loop, HITL），而不是取代人类。 +LabNote 是你不必反复重建的那个版本。它让每个会话从同一个地方开始，提供一套小而明确的方式来留下工作，并让记录在你稍后回来或切换工具时仍能看得懂。 -2. **确定性入口** - 每个 AI 都从同一个入口区进入，阅读同一套规则，并遵循同一条路径。 +你不必使用它的每一个部分。需要审阅、溯源信息、可审计性，或需要干净地把工作交给下一环节时，更周密的路径已经在那里。在此之前，它可以安静地放在你已有工具的旁边。 -3. **受限操作（bounded action）** - AI 获得固定目标、允许路径、停止条件和询问节点。 +## 为什么 LabNote 保持简洁 -4. **工作可追踪** - 文档、批评意见、标签、决定、交接和签退记录都会留下清楚的轨迹。 +LabNote 不会把发生的一切自动保存下来。不是每条消息都应成为长期项目记录。相反，它保存你选择继续带往后面的内容：来源材料、决定、批评意见、交接和下一步。 -5. **增长而不臃肿** - LabNote 可以逐步长成更丰富的工作流，而不要求安装本地守护进程、数据库、模型或常驻仓库的智能体。 +这样会让台账更小、更清楚、更便于检查。后续会话只需沿着有用的轨迹前进，不必无谓地重读、重新总结或加载上下文。 -## 下一步看哪里？ +这些轨道不会让模型更聪明，也不能让模型永远正确。它们让日常协调工作变得更小、更清楚：已知的入口、有限的阅读路径、明确的写入目标，以及应当停止并询问的规定位置。 -LabNote 是工作台账（ledger），不是仓库（warehouse）。工作区应聚焦在体积小、可以直接检查的记录，以及经过批准、指向更重材料的引用。不要在 LabNote 中存放凭证、私钥或 token。 +LabNote 本身不需要： -| 区域 | 需要做什么 | 阅读 | -| --- | --- | --- | -| **开始** | 设置 LabNote | [`docs/ACQUISITION.md`](docs/ACQUISITION.md) | -| **开始** | 入门流程 | [`docs/quickstart.md`](docs/quickstart.md) | -| **开始** | 给 AI 起始说明 | [`AI_ENTRYPOINT.md`](AI_ENTRYPOINT.md) | -| **工作** | 做一次常规投递 | [`lobby/ROUTINE_DEPOSIT_QUICKSTART.md`](lobby/ROUTINE_DEPOSIT_QUICKSTART.md) | -| **工作** | 理解访客会话（visitor）身份 | [`docs/visitor_lobby_model.md`](docs/visitor_lobby_model.md) | -| **工作** | 在会话之间路由消息 | [`docs/message_routing_model.md`](docs/message_routing_model.md) | -| **工作** | 审阅回复或决定 | [`docs/review_workflow.md`](docs/review_workflow.md) | -| **参考** | 分支与 PR 规则 | [`docs/BRANCH_HYGIENE.md`](docs/BRANCH_HYGIENE.md) | -| **参考** | 登记库（registry）的记录与路径 | [`docs/REGISTRY_RECORDS.md`](docs/REGISTRY_RECORDS.md) | -| **参考** | 存储规则 | [`docs/storage_policy.md`](docs/storage_policy.md) | -| **参考** | 文档与二进制文件的投递规则 | [`docs/DOCUMENT_DEPOSIT_POLICY.md`](docs/DOCUMENT_DEPOSIT_POLICY.md) | +- 守护进程； +- 数据库； +- 托管运行时； +- 常驻仓库的智能体； +- 模型 API 密钥。 -详细运行规则见上述对应文件，此处不再重复。若简体中文说明与英文基准文件在协议含义上冲突，以英文基准文件为准。 +不演“共享记忆”这出戏。 -## 致谢 +只有普通文件、清晰路径，以及由人掌握的决定权。这份台账就在文件里，你可以自己检查。 -CapstanAI - LabNote 由人类主导。近期的语言层和验证工作得到了以下 AI 协作者的协助： +[了解 LabNote 如何与上下文文件和模型记忆并存](docs/WHY_LABNOTE.md)。 -- **OpenAI Codex** — 提供实施支持、验证设计和仓库审阅。 -- **DeepSeek** — 提供独立的简体中文语言、文化和对抗性审阅。 +## 它在哪里最值得一用 -这些致谢说明的是辅助性贡献。人类维护者保有作者身份、决策权和发布控制权。 +**研究、调查和个案工作** +把来源、依据材料、不同解释、时间线和决定保存在一起，并跨不同会话延续。 -## 状态 +**软件与技术工作** +在不同工具和智能体之间传递计划、实现说明、测试、审阅和交接。 -当前最新发布的框架版本：`v0.3.2 - Locale Route Freshness`。 +**写作、学习和创作项目** +保留提纲、草稿、批评意见、未决问题和编辑选择，无需每次都重建背景。 -这个补丁版本： +**长期或暂停的项目** +数周或数月后回来时，给一个新的 AI 一个可靠的位置，让它找到真正重要的内容和下一步。 -* 修正了访客会话实际路由中已过时的 zh-CN CSV 指导； -* 使 zh-CN 访客会话入口模型与基准的路由和登记规则保持一致； -* 增加了 CI 检查，用于保护本地化中的关键不变项，并识别未经审阅的成对英文来源变更； +题材其实不太重要。只要工作会持续超过一次聊天、会在工具之间流动，或仅靠记忆重建会很麻烦，LabNote 就能为它提供一条可长期依靠的主线。 -本版本不改变 Registry Contract v1、schema、运行时路径、机器字段、状态值、权限、生成视图或共享协议。 +## 我为什么做它 -`v0.3.0 - Registry v1 and Simplified Chinese Revision` 仍是这些补丁版本所基于的功能版本基线。 +说实话，我受够了一件事：项目里的想法还没琢磨明白，聊天的上下文窗口倒先塞满了。 -以后，CapstanAI 可能会发展出更丰富的确定性层，以及 relay、vault 和 protocol 等模块。LabNote 则从最简单、真正有用的工作台账开始。 +交接说明有所帮助，但每次重新综合都可能漏掉细节，甚至漏掉整条思路。想法会出现在不相关的聊天里，或出现在与另一个模型的对话中。进度报告、批评意见和提醒不断增加，有时项目散乱到我失去线索，只能从头开始。 -```text -工作包、溯源、回复与决定 -``` +项目区域、画布和分支聊天都能帮上忙，但我想要更简单的东西：让一个 AI 会话自然地流进下一个，而不把项目变成庞大的“第二大脑”。 -## 格言 +因此有了 LabNote。它最初是为了解决我糟糕的项目记忆，后来成了我的日常工具。现在，大多数有一定规模的项目我都会为它建立一个；那些搁置数月的旧工作，我也会借它重新整理并继续推进。 -```text -留意间隙，标记交接。 -Mind the gap. Mark the crossing. -``` +归根结底，它让 AI 成为低负担的档案员和协调员：告诉它读什么、把东西放在哪里、记录什么，以及什么时候该停下。 ## 许可证 -Apache License 2.0. \ No newline at end of file +[Apache License 2.0](../../LICENSE) · [CapstanAI LabNote 源仓库的发布记录与版本历史](https://github.com/Wonderforge-Lab/CapstanAI-LabNote/releases) + +> **Mind the gap. Mark the crossing.** diff --git a/locales/zh-CN/docs/PROJECT_PROVENANCE.md b/locales/zh-CN/docs/PROJECT_PROVENANCE.md new file mode 100644 index 0000000..9b59556 --- /dev/null +++ b/locales/zh-CN/docs/PROJECT_PROVENANCE.md @@ -0,0 +1,31 @@ +# 项目溯源记录 + +CapstanAI LabNote 由人类主导。人类维护者保留作者身份、决定权、仓库控制权和发布权。 + +项目在开发、实现、测试、审阅、本地化和文档工作中，也使用了受管理的 AI 协助。下表透明说明这种协助；它不会把责任从人类维护者身上转移出去。 + +## 受管理的 AI 协助 + +| 贡献者 | 在项目中的角色 | +| --- | --- | +| ChatGPT（“Cal”） | 项目设计、文档开发、实现协调和审阅 | +| Claude | 独立批评、测试和编辑审阅 | +| OpenAI Codex | 仓库实现、验证设计和技术审阅 | +| Claude Code | 实现和测试协助 | +| DeepSeek | 简体中文翻译协助，以及语言、文化和对抗性审阅 | + +## 这份记录的含义 + +这是项目范围内的溯源记录。它说明了在项目开发和审阅期间使用过的协助类型。 + +它不是雇佣记录、代理关系、作者身份转移，也不表示任何 AI 系统对项目决定、仓库变更或发布承担责任。 + +## 修订记录 + +具体发布、拉取请求、审阅记录和修订说明会保留各自的局部历史。它们可以写明与该次修订相关的工作或审阅，而不必重复这份完整的项目范围记录。 + +## 本地化 + +英语是基准协议语言。本地化文档可以用自然的本地表达说明同一项目事实，同时保留相关路径、机器字段和协议边界。 + +[返回文档索引](README.md)。 diff --git a/locales/zh-CN/docs/README.md b/locales/zh-CN/docs/README.md new file mode 100644 index 0000000..96ff2d6 --- /dev/null +++ b/locales/zh-CN/docs/README.md @@ -0,0 +1,30 @@ +# CapstanAI LabNote 文档 + +仓库 README 用于帮助你判断 LabNote 是否适合自己的工作，并取得一个副本。本页帮助你在拥有副本后，找到所需的准确指导。 + +[English](../../../docs/README.md) | **简体中文** + +## 开始 + +- [选择设置方式](ACQUISITION.md) —— 将仓库用作 GitHub 模板、在本地克隆或复制，或请 AI 引导你选择它能够支持的最安全方式。 +- [按首次使用流程开始](quickstart.md) —— 在复制出的工作区中开始，无需先学完整个结构。 +- [让 AI 从入口开始](../../../AI_ENTRYPOINT.md) —— 供 AI 会话在 LabNote 内工作时使用的受控阅读路径。 + +## 工作 + +- [进行常规投递](../../../lobby/ROUTINE_DEPOSIT_QUICKSTART.md) —— 以最小但有用的记录，把一项普通工作放到正确路径中。 +- [理解审阅与决定](review_workflow.md) —— 查看回复如何成为经过审阅的结果。 +- [理解访客会话与会话身份](visitor_lobby_model.md) —— 了解实际工作区中的会话记录代表什么。 +- [理解消息路由](message_routing_model.md) —— 遵循会话或角色之间的消息模型。 +- [理解登记记录](REGISTRY_RECORDS.md) —— 查看持久登记记录的用途。 + +## 保持工作区安全 + +- [阅读存储策略](storage_policy.md) —— 让台账保持精简、可检查，并符合工作区可见性的要求。 +- [阅读文档投递策略](DOCUMENT_DEPOSIT_POLICY.md) —— 处理文档和二进制文件，避免把 LabNote 变成普通文件堆。 +- [阅读安全说明](../../../SECURITY.md) —— 不要把凭证、token、密钥或无意暴露的私密材料放入错误的工作区。 + +## 了解项目 + +- [LabNote 与上下文文件和模型记忆](WHY_LABNOTE.md) —— 了解可见的项目台账如何与这两种 AI 支持方式互补。 +- [项目溯源记录](PROJECT_PROVENANCE.md) —— 了解人类维护者与受管理 AI 协助如何参与了本项目。 diff --git a/locales/zh-CN/docs/WHY_LABNOTE.md b/locales/zh-CN/docs/WHY_LABNOTE.md new file mode 100644 index 0000000..0867348 --- /dev/null +++ b/locales/zh-CN/docs/WHY_LABNOTE.md @@ -0,0 +1,27 @@ +# LabNote 与上下文文件和模型记忆 + +LabNote 将经过选择的项目连续性保存在可见文件中。它与模型的上下文窗口和仓库指令文件互补，而不是取代它们。 + +## 模型上下文不等于项目连续性 + +更长的上下文窗口可以帮助模型在一次会话中阅读更多内容。但它本身不会决定什么应在会话结束后继续重要，不会向下一个工具说明为什么做出了某个决定，也不会自动留下人类可以检查的选择性记录。 + +LabNote 不会扩展模型原生记忆，也不会悄悄捕获你的对话。记录由人类或 AI 会话有意写入。因此，这条轨迹可以被检查、修正、审阅，并带到另一个工具中。 + +上下文长度很有用。项目连续性是另一项工作。 + +## 上下文文件设定本地规则 + +`AGENTS.md` 或 `CLAUDE.md` 之类的文件很适合向 AI 说明一个仓库：重要文件在哪里、怎样运行测试，以及有哪些本地规则。 + +LabNote 与它们互补。它的职责是路由持续进行的项目工作：会话从哪里开始、应该读什么、可以把工作留在哪里、如何审阅这些工作，以及何时应停止并询问。 + +上下文文件告诉 AI 它身处什么样的仓库。LabNote 给它一条穿过正在进行的工作的路径。 + +## 使用最小但足够的量 + +如果一次性答案已经足够，就使用手边最好的工具并继续推进。如果普通文件夹已经足够，就使用普通文件夹。 + +当你希望有一条可见的工作轨迹，能随项目成长而不变成隐藏服务或“第二大脑”时，再使用 LabNote。从真正能帮助下一次会话的最小记录开始。 + +[返回文档索引](README.md)。